Mack Institute for Innovation Management - Networks, Social … · 2020-01-06 · Singh and Phelps:...

Information Systems ResearchArticles in Advance, pp. 1–22ISSN 1047-7047 (print) ó ISSN 1526-5536 (online) http://dx.doi.org/10.1287/isre.1120.0449

©2012 INFORMS

Networks, Social Influence, and the Choice AmongCompeting Innovations: Insights from Open

Source Software Licenses

Param Vir SinghDavid A. Tepper School of Business, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213,

[email protected]

Corey PhelpsHEC Paris, 78351 Jouy-en-Josas, 15213 France, [email protected]

Existing research provides little insight into how social influence affects the adoption and diffusion of com-peting innovative artifacts and how the experiences of organizational members who have worked with

particular innovations in their previous employers affect their current organizations’ adoption decision. Weadapt and extend the heterogeneous diffusion model from sociology and examine the conditions under whichprior adopters of competing open source software (OSS) licenses socially influence how a new OSS projectchooses among such licenses and how the experiences of the project manager of a new OSS project with par-ticular licenses affects its susceptibility to this social influence. We test our predictions using a sample of 5,307open source projects hosted at SourceForge. Our results suggest the most important factor determining a newproject’s license choice is the type of license chosen by existing projects that are socially closer to it in its inter-project social network. Moreover, we find that prior adopters of a particular license are more infectious in theirinfluence on the license choice of a new project as their size and performance rankings increase. We also findthat managers of new projects who have been members of more successful prior OSS projects and who havegreater depth and diversity of experience in the OSS community are less susceptible to social influence. Finally,we find a project manager is more likely to adopt a particular license type when his or her project occupiesa similar social role as other projects that have adopted the same license. These results have implications forresearch on innovation adoption and diffusion, open source software licensing, and the governance of economicexchange.

Key words : open source software license; social networks; innovation adoption and diffusion; social influenceHistory : Shaila Miranda, Senior Editor; Ramnath Chellappa, Associate Editor. This paper was received on July

17, 2009, and was with the authors 23 months for 3 revisions. Published online in Articles in Advance.

1. IntroductionThe adoption of innovative artifacts—such asnew ideas, products, technologies, processes andpractices—by members of a social system and theirdiffusion through such systems has fascinated schol-ars from a variety of disciplines and research tradi-tions for decades (Rogers 2003, Phelps et al. 2012).Information Systems (IS) scholars, in particular, havesought to explain the adoption and diffusion of avariety of information technologies (IT) and IT-relatedpractices (e.g., Hardgrave et al. 2003, Jeyaraj et al.2006, Susarla et al. 2012). Although the vast major-ity of IT adoption research and much of the broaderadoption and diffusion literature employs a rationaldecision-theoretic framework and models adoptionas a straightforward cost-benefit analysis (Fichman2004, Hall 2004, Rogers 2003), substantial researchshows adoption decisions often involve profounduncertainty about costs and benefits (Rogers 2003).

Facing such uncertainty, potential adopters typicallyturn to prior adopters as socially influential referentsfor guidance in determining the appropriate adoptionchoice (e.g., DiMaggio and Powell 1983).An emerging theoretical and estimation framework

that incorporates spatial and temporal heterogeneityin modeling the social influence of prior adopters isthe heterogeneous diffusion model (Strang and Tuma1993). This framework decomposes social influence interms of three classes of factors: the susceptibility ofa potential adopter to social information about theinnovation, the infectiousness of information about theinnovation from prior adopters, and the social prox-imity between prior and potential adopters (Greve2005). Although research has used this frameworkto explain, inter alia, the adoption and diffusion ofparticular corporate governance practices (Davis andGreve 1997), prescription drugs (Strang and Tuma1993), and business strategies (Greve 1998), its use by

1

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.

Published online ahead of print November 8, 2012

Singh and Phelps: Networks, Social Influence, and the Choice Among Competing Innovations2 Information Systems Research, Articles in Advance, pp. 1–22, © 2012 INFORMS

IS scholars is rare, despite calls for doing so (Fichman2004). This research suggests the extent to whichprior adopters socially influence potential adoptersdepends on the number and characteristics of prioradopters as sources of information about an innova-tion, characteristics of potential adopters as recipientsand interpreters of this information, and the socialproximity between them. Social influence also variesover time as greater or fewer actors adopt the inno-vation and become more or less socially proximate topotential adopters (Greve 2005).Despite its contributions to an understanding of

how and why innovative artifacts are adopted anddiffused, research employing a heterogeneous diffu-sion framework, as well as the broader innovationadoption and diffusion literature, is limited in twoimportant respects. First, the vast majority of researchrestricts its attention to a single innovation (Strangand Soule 1998). However, individuals and organiza-tions are often confronted with multiple, substitutableinnovations that compete for adoption. This menu ofoptions increases the complexity of the adoption deci-sion and is therefore likely to increase the uncertaintya potential adopter faces, potentially amplifying therole social influence plays in adoption and diffusion.Extant research provides little insight into how, when,or why social influence affects the adoption and dif-fusion of competing artifacts (Strang and Soule 1998).Moreover, the heterogeneous diffusion model has yetto be applied to explain the adoption and diffusion ofmultiple, competing innovations.Second, existing adoption and diffusion research

typically assumes that potential organizational adopt-ers are exposed to an innovation through mass mediaand direct and indirect social ties to prior adoptersrather than direct experience with the innovation.However, because organizations are collections ofindividuals, who often move between organizationsas they change jobs, employees can gain experiencewith and knowledge about particular innovationsadopted by their current organization and transferthis experience to other organizations when theymove (Song et al. 2003). As a result, the natureof the experience employees have had with inno-vations adopted by their previous employers mayinfluence how their new organizations evaluate theinnovations, making them more or less sensitiveto the social influence of prior adopters. Althoughinterfirm employee mobility is a primary mecha-nism by which organizations learn from and influ-ence one another (Song et al. 2003), research doesnot consider when or how the experiences of orga-nizational members who have worked with partic-ular innovations in their previous employers affecttheir current organizations’ adoption of such inno-vations. Little research has examined the characteris-tics of potential organizational adopters, such as their

employees’ prior exposure to particular innovations,which affect their sensitivity to the choices of prioradopters (Wejnert 2002).This study seeks to address these important limi-

tations of innovation adoption research employing aheterogeneous diffusion framework and the broaderadoption and diffusion literature. We do so by inves-tigating the conditions under which prior adoptersof competing innovations socially influence how apotential adopter chooses among such innovationsand how the experiences of members of a poten-tial organizational adopter with particular compet-ing innovations affects its susceptibility to this socialinfluence. Of particular importance to IS research,we investigate this question by studying the adop-tion of particular types of open source software (OSS)licenses by new open source projects.The OSS context is an ideal setting to investigate

our research question for several reasons. First, theOSS licensing framework was an innovative depar-ture from previous legal mechanisms to promotecooperation and beneficial exchange among actors(i.e., software developers and users) with divergentincentives (Demil and Lecocq 2006). Within the broadOSS framework, there is a variety of specific licenses.Because managers of new OSS projects choose alicense at the inception of the project, specific types ofOSS licenses represent discrete, competing innovativelicensing practices that are at risk of being adopted bynew OSS projects. Second, project managers face sub-stantial uncertainty in choosing an appropriate licensebecause of the novelty and large number of licenses(Rosenberg 2000) and the challenge of predicting howtheir choice of a particular license will affect devel-opers’ incentives to join their projects (Shah 2006).This uncertainty is exacerbated by the fact that thechoice of an OSS license is a one-shot, largely irre-versible decision made by licensors who are typicallysoftware developers with little or no legal exper-tise (McGowan 2001). Given this uncertainty, man-agers of new OSS projects are likely to considerexisting projects as social referents that guide theirlicense choice (DiMaggio and Powell 1983). Third,new OSS projects are often initiated by developerswho have previously worked on other OSS projects(Hahn et al. 2008), allowing them to gain experiencewith the licenses used by these projects. These devel-opers carry this experience with them to their newprojects, which may influence how susceptible theirnew organizations are to the social influence of thelicense choices of extant projects. Moreover, in theOSS setting the project manager is the sole mem-ber of the project when the license choice is made,allowing for a clear, direct link between employeeprior experience and organizational adoption choice.A final reason to study OSS license choice is that

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.

Singh and Phelps: Networks, Social Influence, and the Choice Among Competing InnovationsInformation Systems Research, Articles in Advance, pp. 1–22, © 2012 INFORMS 3

although research shows the choice of license by OSSprojects can affect their performance (Comino et al.2007, Stewart et al. 2006), OSS licensing research typ-ically examines the choice between open and closedsource licenses rather than the choice among OSSlicenses. Only two studies have examined the deter-minants of OSS project license choice (Lerner andTirole 2005b, Sen et al. 2009). Both studies, however,ignore the potential social influence of prior adoptersof particular licenses on a project manager’s licensechoice and how this influence may vary over timeand by a project manager’s social proximity to estab-lished OSS projects, characteristics of these projects,and the manager’s previous OSS experience. This is asurprising and substantive limitation of this researchgiven the substantial uncertainty managers of newOSS projects face in choosing a license, which sug-gests they may be socially influenced by the licensechoices of existing projects.We adapt and extend the heterogeneous diffusion

model to accommodate multiple, competing inno-vations. To do so, we investigate each dimensionof this model—infectiousness, social proximity, andsusceptibility—and derive predictions related to eachdimension. We incorporate a novel source of influenceon a potential adopter’s susceptibility to the socialinfluence of prior adopters—namely, the experiencesof organizational members who have worked withparticular innovations in other organizations.We test our predictions in a sample of 5,307 OSS

projects hosted at SourceForge. After controlling forfactors shown to affect OSS license choice (Lerner andTirole 2005b), our results suggest the most importantfactor determining a new project’s license choice isthe type of license chosen by existing projects that aresocially closer to it in its inter-project social network.Moreover, we find that prior adopters of a particularlicense are more infectious in their influence on thelicense choice of a new project as their size and perfor-mance rankings increase. We also find that managersof new projects who have been members of more suc-cessful prior OSS projects and who have greater depthand diversity of experience in the OSS community areless susceptible to social influence. Finally, we find aproject manager is more likely to adopt a particularlicense type when his or her project occupies a simi-lar social role as other projects that have adopted thesame license.This study contributes to the innovation adoption

and diffusion literature by addressing important lim-itations of the heterogeneous diffusion frameworkand to the literature on open source software bybeing the first study to explore when and howsocial influence from existing OSS projects affectsa new project’s license choice. This study also has

substantive implications for understanding the ori-gins and influence of the social institutions that gov-ern economic exchange.

2. Open Source Software2.1. Open Source Software Development ProcessAll OSS projects follow a similar process. An “initi-ating developer” begins a project by working on anidea and then hosts the source code and invites otherdevelopers to contribute. Developers volunteer to per-form specific tasks and collaborate as a team, incorpo-rating their individual creations into a single body ofsource code. Once an executable version of the soft-ware is developed, it is released for testing and feed-back. The software evolves as new features are added,existing features are modified, and bugs get fixed.The process involves the sharing of ideas and jointproblem solving that fosters social bonds among col-laborators. Given the small number of developers typ-ically involved in OSS projects (Krishnamurthy 2002)and the frequency and intensity of their interactionsover time, the social ties among them can be quitestrong (Hahn et al. 2008, Singh et al. 2011a, Singh andTan 2010, Singh 2010). Because OSS projects stimulatethe formation of social ties among teams of develop-ers and because developers often work on multipleprojects, a social network is produced that directlyand indirectly connects developers participating inthe broader OSS community. Although projects cre-ate ties among developers, projects also become con-nected by sharing common developers (Grewal et al.2006). We examine the influence of this latter inter-project social network in this study.

2.2. Open Source Software LicensesTo be characterized as “open source,” software mustbe offered under a license that satisfies several con-ditions.1 Both the Free Software Foundation and theOpen Source Initiative approve OSS licenses. OSSlicenses differ in the extent to which they restricthow users may use and modify the software. Atone extreme are highly restrictive licenses, such as theGNU general public license (GPL), and at the otherextreme are permissive licenses, such as the BerkeleySoftware Distribution (BSD) license. Highly restrictivelicenses differ from permissive licenses in two keyways (de Laat 2005):(1) They require that, when modified versions of

the program are distributed, the source code must bemade generally available. This provision is called the“copyleft” clause.

1 The lists of conditions and the rationale behind them areavailable at http://www.opensource.org/docs/definition.php andhttp://www.gnu.org/philosophy/free-sw.html.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


(2) They prohibit the software to be mingled withother software that does not use the same license.This provision is called the “viral” or the “reciprocal”clause.The highly restrictive licenses were the first free

software licenses. The most famous of these, GPL, wasauthored by Richard Stallman, an early proponentof OSS and the initiating developer of a free oper-ating system, GNU. The copyleft and viral clauseswere designed to protect the software from beinghijacked by proprietary software developers. Theseclauses require that any modification or derivation ofthe software has to be offered under GPL, makingGPL’d software less attractive to commercial actors.The viral clause restricts the software from exploitingcomplementarities with other software, reducing itsappeal to both contributors and users. Although theBSD license makes software attractive for commercialuse because it allows modified or derivative works tobe kept private, it does not protect the software frombeing hijacked. Stallman authored the Lesser GeneralPublic License (LGPL) as a compromise between thehighly restrictive GPL and the permissive BSD license(Stallman et al. 2002). The LGPL includes the copyleftbut not the viral provision. Lerner and Tirole (2005b)refer to LGPL-type licenses as restrictive. Any OSSlicense can be categorized as permissive, restrictive,or highly restrictive.

3. Prior Research on the Choice ofOpen Source License

Few studies have examined the determinants of OSSproject license choice (Lerner and Tirole 2005b, Senet al. 2009). Lerner and Tirole (2005b) model the licen-sor’s problem as an optimizing balancing act betweenchoosing a more restrictive license to attract morecontributing developers and adopting a permissivelicense to preserve her ability to commercialize thesoftware. Conditional on the exogenous characteris-tics of the software development project, the man-ager chooses the license that maximizes her expectedbenefits from the project given her evaluation of theexpected response by potential developers. Consis-tent with their expectations, Lerner and Tirole (2005b)found that characteristics of an OSS project–such asits intended audience (e.g., end users), applicationgenre (e.g., gaming), operating system (e.g., Linux),and user interface (e.g., GUI)–determine its choiceof license (Lerner and Tirole 2005b). In contrast tothe focus on project characteristics, Sen et al. (2009)examine how the choice of OSS license type can beexplained by the intrinsic and extrinsic motivationsand the attitudes of the project manager. They showedthat project managers who were motivated by theproblem-solving challenges of OSS projects preferred

moderately restrictive licenses, whereas those moti-vated by peer recognition preferred unrestrictivelicenses. They also found that, when choosing alicense, project managers are more concerned withthe ideological principle that all OSS should be ableto be freely redistributed than they are with end-users’ rights and that licensors prefer licenses that arealigned with these attitudes.Despite the insights this research provides into

understanding OSS license choice, neither study con-siders the potential social influence of prior adoptersof particular licenses on a new project’s licensechoice. This represents an important limitation of thisresearch because a manager of a new OSS project islikely to be socially influenced by existing projects’licensing choices given the substantial uncertainty heor she faces in choosing a license (as described above).Indeed, we visited many online forums in whichdevelopers queried one another about which licenseto choose for their projects and observed substantialconfusion and uncertainty on this topic. The two post-ings below are representative of what we observed:The number of licenses is INSANE, I may as well just

write my own and add it to the list ( joking 0 0 0mostly)I have a few open source apps and I think they are probablynot licensed as to what I want, but I can’t seem to find agood source to get a suggestion, and I don’t want to readall of them.2I wish there was just a spreadsheet of “This license has

this feature.” I’m on the verge of releasing something andit’s very hard to pick the right license.3To explore the role social influence plays in the

choice of open source license, we build on andextend the heterogeneous diffusion model (Strangand Tuma 1993).

4. Heterogeneous Diffusion ModelThe heterogeneous diffusion model was developedto explain how social context influences an actor’sadoption of an innovation and its diffusion withina population of actors (Strang and Tuma 1993). Thismodel allows for spatial heterogeneity in the influenceof prior adopters on individual adoption behaviorand accommodates temporal heterogeneity by allow-ing the influence of prior adopters and social prox-imity to vary over time (Strang and Tuma 1993).In this model, the extent to which an actor’s socialcontext influences its adoption behavior depends onthree categories of specific explanatory factors: infec-tiousness, social proximity, and susceptibility (Strang and

2 Accessed on 10/30/2011 from http://stackoverflow.com/questions/450378/how-to-choose-an-open-source-license-for-an-app.3 Accessed on 10/30/2011 from http://programmers.stackexchange.com/questions/120308/how-can-i-compare-and-contrast-open-source-licenses.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Tuma 1993). The infectiousness of a prior adopter ofan innovation refers to how influential informationgenerated about its actions is for potential adopters,which is a function of characteristics of the prioradopter such as its size, performance, or status (Greve2005). The social proximity of the source to a poten-tial adopter describes how easily information is trans-mitted between them, based on their social distancefrom each other (Greve 2005). The susceptibility ofa potential adopter refers to how open, receptive,or sensitive it is to being influenced by informa-tion available about the innovation and depends oninherent attributes that affect its motivation and abil-ity to adopt the particular innovation (Greve 2005).The mechanism by which prior adopters influencepotential adopters is the transmission of informa-tion between them, either through direct and indirectsocial ties, or through potential adopters observingprior adopters (Strang and Soule 1998). The modelalso specifies that an actor has intrinsic propensities toadopt an innovation based on its own inherent char-acteristics, independent of social context. We considerintrinsic characteristics affecting adoption as controlvariables that provide a baseline model.Because the OSS licensing framework represents

a novel set of contractual practices for protectingintellectual property (Demil and Lecocq 2006), wecharacterize types of OSS licenses as discrete and sub-stitutable innovative practices that are at risk of beingadopted by new OSS projects. We focus on the adop-tion of specific categories of licenses rather than indi-vidual OSS licenses. Although there are over 40 OSSlicenses, they are small variations on two underly-ing themes—whether they contain copyleft and/orviral clauses (de Laat 2005, Lerner and Tirole 2005b,Stewart et al. 2006)—and thus represent three cate-gories (types) of OSS licenses: unrestrictive (neitherclause is present), restrictive (only the copyleft clauseis present), and highly restrictive (both clauses arepresent) (Lerner and Tirole 2005b). Because these cat-egories capture the essential differences among OSSlicenses (de Laat 2005), each type represents a proto-typical OSS license. Given the uncertainty surround-ing these complex artifacts, potential licensors willtend to simplify their licensing decisions by focusingon prototypical licenses because these represent theessential and salient differences among the many indi-vidual licenses (Kahneman et al. 1982). Indeed, onlinetools (such as OSS Watch and Three.org) and arti-cles (e.g., Niiranen 2009) designed to help developerschoose OSS licenses typically explain and prescribethe licenses in terms of these prototypical charac-teristics rather than peripheral features. When aninnovative practice is complex and uncertain, its pro-totypical characteristics rather than peripheral fea-tures drive its adoption (Ansari et al. 2010). Next, we

define the type of social network in the OSS settingthrough which we expect social influence to operate.We then develop hypotheses linking variables associ-ated with each dimension of the heterogeneous diffu-sion model to the likelihood a new OSS project willadopt a particular license type.

4.1. Inter-Project Social NetworkThe nature of the development process in OSS leadsto the emergence of affiliation networks (Grewal et al.2006). An affiliation network is a two-mode networkbecause it consists of actors connected by their par-ticipation in common events and events that are con-nected by common actors (Wasserman and Faust1994). An affiliation network therefore represents twodifferent types of one-mode networks: an interactornetwork and an inter-event network (Wasserman andFaust 1994). In the OSS setting, the actors are individ-ual developers and the events are projects. Developershave social ties with one another as a result of work-ing together on the same project and projects are con-nected to one another as a result of sharing commondevelopers (Grewal et al. 2006). Rather than focus onthe inter-developer social networks that result fromOSS affiliation networks, we focus on inter-projectnetworks as the appropriate social network. An OSSproject administrator initiates a project and is the soledeveloper on the project at the time of project regis-tration and license choice. Thus, at the time of licensechoice, the project and administrator (i.e., licensor) areone and the same. Our theory suggests an adminis-trator is influenced by his or her social proximity to,and characteristics of, other projects (rather than indi-vidual developers). Thus, an administrator’s relevantsocial network at the time of license choice consistsof the social ties the administrator-as-project has topreviously established OSS projects.

4.2. Social ProximitySocial proximity refers to the social distance betweentwo actors in a social network and determines howeasily information is transmitted between them andthe relevance of this information (Coleman et al.1966). A socially proximate actor provides an influ-ential frame of reference by which a focal actorevaluates and interprets information (Leenders 2002).Two approaches to conceptualizing social proximityin a network exist, each with its own causal mech-anism linking proximity with social influence. Thefirst approach—social cohesion—defines proximity interms of the number, length, and strength of thepaths that connect actors in a network (Marsden andFriedkin 1993). The second approach—equivalence—defines proximity in terms of the similarity of twoactors’ profiles of network relations (Marsden andFriedkin 1993).

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


4.2.1. Social Cohesion. The social cohesionapproach defines social proximity in terms of thenumber, length, and strength of the paths thatconnect actors in a network (Marsden and Friedkin1993). We focus on social distance (i.e., path length)as the primary dimension of social cohesion. Thesimplest form of cohesion is when two actors, suchas a potential and prior adopter of an innovation,share a direct social tie. Directly connected actorscommunicate and share information with each othermore frequently and with greater fidelity than indi-rectly connected actors (Burt 1982). Direct ties areconduits for the communication of rich, personalizedinformation, which tends to be more influential thanimpersonal information sources (Rogers and Kincaid1981). The volume and fidelity of information decaysas the number of links indirectly connecting actorsincreases (Shannon 1949), making indirectly con-nected actors less socially influential on a potentialadopter than direct contacts (Burt 1982). Research insocial psychology suggests involvement in sharedactivities provides opportunities for social cohesionto develop and that shared attitudes develop fromsocial cohesion (Homans 1961). Faced with an uncer-tain situation, such as the adoption of an innovation,individuals discuss it with their proximate peers anddevelop a consensual normative understanding of theassociated costs and benefits (Rogers 2003). Social tiesprovide detailed, personalized and more persuasiveinformation on costs and benefits of adoption thangeneral information sources (Rogers 2003). Discus-sions with prior adopters of an innovation buildsocial pressures on the potential adopter to adopt theinnovation when faced with an opportunity to do so(Rogers and Kincaid 1981). Social pressure increaseswith social cohesion and hence a potential adopteris more likely to adopt an innovation that has beenadopted by his or her most proximate peers. Researchhas found that shared attitudes and behavior developamong people or organizations that are connectedthrough direct communication channels (Colemanet al. 1966, Davis and Greve 1997, Haunschild 1994).In the OSS context, projects on which a licensor has

worked in the past provide greater opportunities forcommunication and thus for social cohesion. Throughher discussions with developers on prior projects,a focal licensor develops a shared understanding ofthe costs and benefits associated with the license typechosen for those projects. Although direct involve-ment in a project provides opportunities to observethe consequences of adopting a license, a licensor mayalso receive useful and persuasive information fromprojects with which she is not involved but has socialties with developers who are. These ties can providethe licensor with detailed, personalized, and persua-sive information on the costs and benefits of the par-ticular license types adopted by these projects. Prior

adopters have experience with a particular licensetype and thus understand it better than a potentialadopter and may communicate their preferences per-suasively via social ties with the focal project admin-istrator. Hence, socially cohesive prior adopters exertsocial pressure on a potential licensor to adopt thesame license type.

Hypothesis 1 (H1). A licensor is more likely to choosea license type that was adopted by other projects to whichhe or she is more closely socially connected (i.e., sociallycohesive).

4.2.2. Role Equivalence. An alternative concep-tualization defines social proximity in terms ofequivalence—the similarity of two actors’ profilesof network relations (Marsden and Friedkin 1993).The social network literature initially conceptualizedequivalence as structural equivalence. Two actors arestructurally equivalent to the extent they have tiesto the exact same other actors (Burt 1976). Researchfindings on the influence of structural equivalenceon actor behavior can, however be interpreted intwo ways (Mizruchi 1993). First, competition betweenactors over the same resources provided by the samealters triggers imitation between socially substitutableactors (Burt 1987). Second, social cohesion because ofdirect ties with the same set of third parties inducessimilar behaviors.Because of these alternative interpretations, we

employ the concept of role equivalence to capturehow competition among actors, such as prior andpotential adopters, results in imitative behavior. Twoactors are role equivalent to the extent they engagein the same kinds of relationships with third par-ties (Mizruchi 1993). For example, two organiza-tions are more role equivalent when they produceor trade in the same products, and therefore havesimilar types of upstream and downstream relation-ships, although not necessarily with the same suppli-ers or buyers (Guler et al. 2002, Winship and Mandel1983). In this vein, equivalence among organizationshas been defined as the extent to which they pro-duce similar products or produce products using sim-ilar technologies and therefore have similar kinds ofvertical relationships (Bothner 2003, Davis and Greve1997, Flingstein 1985). Role equivalence captures thedegree to which two actors occupy similar social rolesand thus serve as common referents for one another,regardless of whether or not they share direct ties orare structurally equivalent (Mizruchi 1993).Because role equivalence does not depend on the

presence (or absence) of a tie between the actorsbeing compared, the social influence effect of roleequivalence on actor behavior is different from theinfluence associated with direct ties. The causal mech-anism linking equivalence and the similarity of actors’

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


behavior is social influence via a process of obser-vational social comparison. Although role equivalentactors do not necessarily share relations with the samethird parties, they compete with one another to retaintheir existing ties because third parties view suchactors as substitutable objects of interaction (Guleret al. 2002). Competition among two equivalent actorsincreases their incentives to monitor and comparebehaviors to ensure neither has an advantage or fallsbehind (Burt 1987). The more equivalent, the moreone actor is likely to adopt an innovation previouslyadopted by another because it may make the othermore attractive as the object or subject of relation-ships, resulting in some alters abandoning the non-adopter in favor of the actor that adopted (Guler et al.2002). Nonadopters monitor and imitate the adoptionbehavior of role equivalent adopters in order to retainexisting relationships and the benefits they provide(Guler et al. 2002). Actors look to equivalent others toidentify appropriate behavior, particularly in contextscharacterized by substantial uncertainty about how tobehave (Burt 1987). The equivalence model of socialproximity highlights symbolic communication amongsocial substitutes rather than direct communicationamong contacts, which a social cohesion perspectiveemphasizes (Leenders 2002). This argument is consis-tent with the emphasis in neo-institutional researchon mimetic isomorphic processes within industries(DiMaggio and Powell 1983). Prior research showsthe extent to which organizations produce similar,competitive products or employ the same technolo-gies are more likely to imitate each other’s adoptionsof innovations (Bothner 2003, Davis and Greve 1997,Flingstein 1985).Based on prior research (Guler et al. 2002, Winship

and Mandel 1983), we define role equivalence as theextent to which OSS projects engage developers andusers in the same technologies. As such, OSS projectsare role equivalent to the extent they employ the sametechnology platform in their development. Indeed,the primary data source we use in this study, Source-Forge, organizes all OSS projects it hosts into com-mon domains or “foundries” based on their commonusage of a technology platform, such as projects thatuse the Perl programming language. Projects in thesame foundry typically target and compete for simi-lar users and for developers with similar skills (Hahnet al. 2008). The extent to which OSS projects competeto maintain relationships with users and developersincreases the extent to which the projects are socialsubstitutes (from the perspective of users and devel-opers). This increasing competition should increasethe incentives OSS projects have to monitor and imi-tate each other’s adoption of practices that can influ-ence the performance of their respective projects, suchas the choice of license. Because license choice affects

OSS project performance and such choice involvesconsiderable uncertainty, a licensor of a new projectshould be particularly sensitive to the license(s) pre-viously adopted by equivalent projects. Compatibil-ity issues associated with an OSS project’s choiceof license may increase the tendency to mimic theprevious licensing decisions of equivalent projects.Complementarities often exist among software prod-ucts and the exploitation of these complementaritiesinfluences a product’s success (Gallaugher and Wang2002). The license that governs the development anddistribution of a software product influences its abil-ity to exploit complementarities with other softwareproducts because such licenses influence the extentto which products are legally compatible and can bemingled and mixed by developers and users (Lernerand Tirole 2002). Licensors may fear the opportu-nity costs of choosing an inappropriate license—onethat reduces its compatibility with other projects—and thus look for guidance on the appropriate choicefrom equivalent projects. In sum, a licensor will bemore likely to monitor and imitate the license choicesof projects in the same foundry.

Hypothesis 2 (H2). A licensor is more likely to choosea license type that role equivalent projects have adoptedmore widely.

4.3. Infectiousness of Prior AdoptersThe infectiousness of a prior adopter of an innova-tion describes how influential the information aboutits actions is for the adoption decision of a poten-tial adopter. Because infectiousness is a property ofinformation generated by prior adopters, its influ-ence depends on whether or not it can reach a poten-tial adopter and the distance it must travel to do so(Strang and Tuma 1993). In other words, the influ-ence of infectiousness depends on the social proximitybetween a prior and potential adopter. Prior researchsuggests prior adopters can be more or less influ-ential as social referents for a subsequent potentialadopter based on characteristics such as their size,performance, or status (Greve 2005). We focus onprior adopter performance as the primary driver ofits infectiousness.4 License adoptions by higher per-forming OSS projects will be more influential on thechoice of license type by new projects for several rea-sons. First, successful projects attract more attention,which leads to more information being available topotential adopters about them (Greve 2005). Second,potential adopters often attribute the level of success

4 Although we focus on a prior adopter’s performance as the pri-mary construct affecting infectiousness, we argue that OSS projectperformance determines its status within the OSS community(Stewart 2005) and we use the number of developers a projectattains (i.e., its size) as a measure of project performance.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


an actor subsequently achieves to their choice of inno-vation (Greve 2005), which suggests potential OSSlicense adopters may attribute a prior adopter’s suc-cess, in part, to its choice of license, making infor-mation about such adopters more influential on theadoption decisions of others. However, such attri-butions may be unnecessary because people oftencopy the behavior of successful or prestigious refer-ents, even when the focal behavior has nothing todo with the referent’s success or prestige (Henrichand Gil-White 2001). Third, whereas the adoption ofan OSS license is characterized by substantial uncer-tainty, innovations adopted by high status organiza-tions are viewed as less uncertain and therefore morelikely to be imitated by others (DiMaggio and Powell1983). In this way, an actor’s status provides anuncertainty-reducing signal about its underling qual-ity and competence, increasing the extent to whichothers attend to and are influenced by its actions(Podolny 2001). Such status signals are particularlyinfluential when recipients find it difficult to searchfor other useful information sources to inform theiradoption decision (Simcoe and Waguespack 2011). Inthe OSS community, the success of a project reflects itssocial status in the community (Stewart 2005). Thus,the adoption of a particular license type by success-ful projects is likely to be imitated by new projects.Because the social proximity of established projectsdetermines the extent to which they will be usedas social referents by a focal project, we expect thelicense types chosen by socially proximate projectswill have a stronger effect on a focal project’s licensechoice when these projects are more successful.

Hypothesis 3 (H3). A licensor is more likely to choosea license type that has been adopted by successful projectsthat are socially proximate.

4.4. Susceptibility of LicensorSusceptibility refers to how much a potential adopteris affected by information about the practices adoptedby others (Greve 2005). Holding constant the infor-mation about prior adopters that reaches potentialadopters, a potential adopter is more susceptiblewhen his or her adoption decision is more sensitiveto and influenced by this information (Greve 2005).Little research has examined the characteristics ofpotential organizational adopters that affect their sen-sitivity to the choices of prior adopters (Wejnert 2002).In particular, research does not consider when or howthe experiences of organizational members who haveworked with particular innovations in other organi-zations affect their current organizations’ adoption ofsuch innovations. Potential adopters differ in theirsusceptibility to the social influence of prior adoptersbased on differences in a potential adopter’s moti-vations to search for new practices and learn from

the actions of others (Greve 1998). In addressing animportant limitation of extant innovation adoptionresearch, we argue that these motivational differenceswill be affected by the nature of the experience keyorganizational members (i.e., the project manager)have had with the innovations in their work on otherprojects that had previously adopted.The success and failure of other OSS projects on

which the licensor of a new project has worked willinfluence her susceptibility. Individuals and organiza-tions are motivated to search for new practices andlearn from others’ efforts when they are dissatisfiedwith their own performance (Cyert and March 1963).Thus, the search for new behaviors or practices is trig-gered by the need to solve the problem of poor per-formance (Cyert and March 1963). Actors simplify theevaluation of their performance by assessing it rel-ative to their aspiration level and by dichotomizingactual performance as “success” or “failure” depend-ing on whether it was above or below their aspirationlevel (March and Simon 1958). An aspiration level is“the smallest outcome that would be deemed satisfac-tory by a decision maker” (Schneider 1992, p. 1053).Aspiration levels are determined by a process ofsocial comparison (Cyert and March 1963, Festinger1954, Greve 1998). Individuals compare themselvesand their organizations with referent others who aresimilar to the focal actor and who therefore serveas a reference group (Greve 2005). A measure of theaggregate performance of the reference group, suchas the average, constitutes the social aspiration levelfor an actor, who compares its own performance tothis level to determine success or failure (Cyert andMarch 1963, Greve 2005). Poor performance triggersthe search for new behaviors and practices, whereassuccess decreases the likelihood of such search. In thecontext of innovation adoption, performance belowan actor’s social aspiration level increases its willing-ness to consider adopting innovations adopted by itsreferent group, whereas success reduces incentives tochange existing behavior and practices (Greve 2005).Poor performing organizations have been found to bemore susceptible to social influence when faced withdecisions to adopt novel practices (Davis and Greve1997, Kraatz 1998).In the OSS context, we expect the extent to which

a licensor of a new project has previously worked onsuccessful or unsuccessful projects will influence hersusceptibility to the social influence of other sociallyproximal projects. As discussed previously, sociallyproximal projects are likely to serve as the referentgroup for the licensor of a new project. Relativelypoor performance in prior projects should increase alicensor’s propensity to search for appropriate prac-tices, which will increase his or her susceptibility tooutside influence. Experience working on successful

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


OSS projects, in contrast, will tend to reduce a licen-sor’s willingness to attend to the adoption deci-sions of other projects. Prior successful experiencealso sends a positive signal about the capabilitiesof the licensor to the community (Podolny 2001,Stewart 2005), which increases the project’s attrac-tiveness to potential contributors, irrespective of thelicense adopted. Such a licensor is less likely to seekhelp from others to determine the appropriate licensetype for his or her new project. Therefore, we expectthat a licensor’s susceptibility to influence from otherprojects decreases with the success of his or her priorOSS projects.

Hypothesis 4 (H4). The effects of social proximity(social cohesion and role equivalence) on the likelihood alicensor will adopt a particular license type will decreasewith the relative success of the other OSS projects on whichthe licensor worked.

The depth and breadth of a licensor’s experiencewith OSS projects will also influence her susceptibil-ity to social influence. The need to learn from others’behavior increases with a potential adopter’s unfa-miliarity with an innovation or the setting in whichit is used (Wejnert 2002). When encountering some-thing previously unknown or out of the ordinary, anindividual begins a process of inquiry (Shultz 1964)and relies upon information and interpretation fromothers to make sense of the situation (Louis 1980).As individuals gain experience in a particular setting,they normally know what to expect of a situation,making them less susceptible to others’ interpreta-tions or behaviors (Louis 1980). Although the adop-tion of an OSS license has uncertain implications,licensors may differ in their familiarity with the OSScommunity and its practices, owing to differences intheir levels of experience. OSS licensors with substan-tial depth of experience on OSS projects will tendto have more knowledge about OSS licenses thanless experienced developers. Greater experience andknowledge increases individuals’ sense of self-efficacyand reduces their incentives to alter their behavior(Bandura 1986). In particular, such individuals are lesssusceptible to social influence in their decision mak-ing processes. Moreover, the accumulation of knowl-edge by individuals about a particular phenomenonincreases their propensity to exploit this knowledgein making decisions and solving problems, ratherthan search for information and solutions outsidetheir expertise, because local search tends to be lesscostly and generates less variable results than exter-nal, exploratory search (Audia and Goncalo 2007).Thus, licensors with more, or “deeper,” experiencewill be less susceptible to social influence.The diversity of a licensor’s experience will also

reduce her susceptibility. Experience with a variety

of different licenses increases a licensor’s familiar-ity with more licenses, reducing the likelihood thelicensor will need to rely on others for informationabout a particular license. In general, diverse experi-ence with a particular phenomenon encourages indi-viduals to consider the phenomenon from a varietyof perspectives, which challenges their fundamentalassumptions and stimulates deeper and richer causalunderstandings about it (Argyris and Schön 1974).This richer understanding fosters healthy skepticismin decision making, reduces the potential for decisionbiases, and improves decision quality (Janis 1972).Thus, licensors who have experience with a greaterdiversity of OSS licenses will posses richer knowl-edge about such licenses and will consequently haveless incentive to seek out or be influenced by externalsources of information. In sum, a licensor with deeperand more diverse experience with OSS licenses willbe less susceptible to social influence in her licensingdecision.

Hypothesis 5 (H5). The effects of social proximity(social cohesion and role equivalence) on the likelihood alicensor will adopt a particular license type will decreasewith the depth and diversity of the licensor’s prior experi-ence in the OSS community.

5. Data and Methods5.1. Sample and Social Network ConstructionTo test our hypotheses we constructed a data set ofOSS projects hosted at SourceForge.net (SF). SF isthe largest hosting site for OSS projects and accountsfor about 90% of all OSS projects. SF provides Webspace as well as services such as emailing facil-ity, discussion forums, CVS5 repository hosting, anddownload servers for OSS projects to organize andcoordinate their development activities. Many stud-ies of OSS have used SF data (e.g., Comino et al.2007; Grewal et al. 2006; Lerner and Tirole 2005b;Singh et al. 2011a, b). We consider all projects regis-tered on SF between its inception in November 1999and December 2003. Data regarding project character-istics and developers were obtained from project web-sites and the SF database. To ensure the developerswhose names are associated with the projects in theSF database actively participated in the projects dur-ing the period under consideration, we matched theircontribution efforts and time of involvement throughCVS log files, project communications, and the projectdocumentation.The hypotheses focus on a licensor’s social net-

work. To test our hypotheses about the influence of

5 Concurrent Versioning System (CVS) is a software tool that storesthe project source code, tracks changes made to the source code,and stores programmers’ comments about the changes they made.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


these social networks, we need to observe them. Con-structing a social network starts with the identifi-cation of the network’s boundary—the collection ofactors who, for analytical purposes, is regarded asa bounded social collective (Marsden 2005). Build-ing on prior research that identifies criteria for estab-lishing network boundaries for empirical research(Laumann et al. 1983), we used participation in afoundry as the network boundary. A foundry includesall projects that use a common technology platform,such as projects that share the Perl programming lan-guage. Using a foundry as a network boundary cri-terion is valid for two reasons. First, projects withinfoundries are technologically similar and, as such,foundries provide a meaningful context for knowl-edge sharing among network members. Participationin a foundry as a network boundary for OSS projectsat SF has been used in related research (Grewal et al.2006, Singh 2010, Singh et al. 2011a). Second, socialrelationships almost always occur among individualswithin a foundry and rarely occur across foundries.We analyzed 2,000 randomly selected developers whoworked on multiple projects and found that onlyabout 4% worked across two or more foundries.In constructing our sample, we considered all

projects that were registered in all 22 foundries atSF between November 1999 and December 2003.6We discarded projects that never showed any devel-opment activity, leaving 29,995 projects. Althoughwe use all of these projects to construct our net-works, we limit our analysis to explaining varia-tion in license adoption to projects that were startedbetween January 2002 and December 2003 (n =211220). Because relationships endure over time, con-structing networks using only relationships formedat a particular point in time would greatly under-state the network’s true connectivity. Data on bothpre-sample relationship formation and relationshipduration are needed to accurately assess networkstructure at a point in time. We assume that rela-tionships formed in the presample period endurethrough the end of the sample period (December2003). The assumption of individual social ties decay-ing after four years is consistent with prior researchon interpersonal affiliation networks (e.g., Cattani andFerriani 2008, Uzzi and Spiro 2005). Because our socialnetwork measures are meaningful for only those newprojects at risk of license adoption that are directlysocially connected to other projects, we remove fromour final analysis all socially isolated projects—i.e.,projects that do not have relationships with other

6 The beginning of this time period represents the beginning ofthe SourceForge data. We ended data collection at the end of 2003because this represents the closest year end to when we began thisresearch project.

sample projects (n = 151913), leaving us with a finalsample of 5,307 projects. As we explain below, weaccount for the possibility that this self-selection intoour foundry-based social networks biases our results.

5.2. Network ConstructionTo assess social proximity, we constructed 22 affiliationnetworks. Each network contains projects assignedto a particular foundry. In these networks develop-ers have social ties with one another as a result ofworking together on the same project and projects arerelated to one another as a result of sharing devel-opers. Because a project’s administrator is responsi-ble for choosing the project’s license and is the onlymember of the project at the time of license choice,the project and administrator are one and the sameat the time of license choice. Our theory suggests anadministrator is influenced by his or her social prox-imity to, and characteristics of, other projects (ratherthan individual developers). Because the unit of anal-ysis is the project, we assessed social proximity amongprojects by projecting each developers-by-project affil-iation network into its respective unipartite (i.e., onemode) inter-project network.

5.3. Dependent VariablesWe expect a licensor of a new project will be morelikely to adopt the type of prototypical license thatprojects to which she is socially proximate haveadopted. Prototypical licenses are defined by whetherthey include copyleft and/or viral clauses becausethese clauses define the essential differences amongOSS licenses (de Laat 2005, Lerner and Tirole 2005b,Stewart et al. 2006). These two license clauses havefundamentally different implications for legally copy-ing, modifying, and distributing software sourcecode; represent fundamental ideological differencesamong proponents of OSS; and have important impli-cations for the success of OSS projects (de Laat2005, Stewart et al. 2006). Accordingly, we categorizelicenses using these clauses. Because a project mayoffer its software under multiple licenses we followLerner and Tirole (2005b) and categorize a projectunder four different categories: (1) all licenses highlyrestrictive or not (ALLHR), (2) some licenses highlyrestrictive or not (SHR), (3) all licenses restrictive ornot (ALLR), and (4) some licenses restrictive or not(SR).7 These categories are not mutually exclusive.

7 A total of 19% of sample projects use multiple licenses. Conse-quently, dropping sample projects with multiple licenses wouldsubstantially misrepresent the true inter-project social networkstructure and bias social cohesion measures. For example, assumea focal project A was directly connected to two other projects(B and C). Assume project B has adopted multiple highly restrictivelicenses and C has adopted a single unrestrictive license. Half ofthe projects to which project A is socially connected have adopted

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


The requirement for a license to be classified as highlyrestrictive is the presence of both copyleft and viralclauses, whereas the presence of only the copyleftclause is needed for it to be classified as restrictive.Thus, all highly restrictive licenses also satisfy therestrictive license requirement. Hence, SHR is a subsetof SR and ALLHR is a subset of ALLR. To compareour results with those of Lerner and Tirole (2005b),we follow them and treat each of these four categoriesas separate dichotomous dependent variables.We estimate four different license choice models.

In model 1, the dependent variable is coded “1” ifall licenses chosen by the focal project were highlyrestrictive and “0” otherwise. The dependent vari-ables for other models were coded similarly. Becauseour theory specifies that a new project is more likelyto adopt a particular type of OSS license when sociallyproximate projects have previously adopted the samelicense type, we do not construct a single dependentvariable that reflects increasing license restrictivenessand model the likelihood a project will adopt a morerestrictive license using, for example, an ordered pro-bit estimator. As robustness checks of our primaryresults, however, we use two alternative constructionsof our dependent variable. We discuss these robust-ness checks in the online supplement (available athttp://dx.doi.org/10.1287/isre.1120.0449).

5.4. Explanatory Variables

5.4.1. Social Cohesion. Social cohesion definessocial proximity in terms of the number, length, andstrength of the paths that connect actors in a network(Marsden and Friedkin 1993). We focus on social dis-tance as the primary dimension of social cohesion.Our measure of social cohesion is based on a mea-sure of social distance in which distance is defined asthe number of network links that exist on the short-est path between two connected actors. To computethis measure, we first calculate the social distanceamong each pair of projects in the foundry-specificinter-project network. The social distance matrix isthen used to calculate nearness among each pairof projects using an exponential decay transforma-tion (Burt 1982). Larger values of nearness representstronger social cohesion between two projects. Socialdistance is calculated from all projects to the focalproject for which there is a path between projects.Because social distance is only meaningful for projectsthat started at SF prior to the focal project, we do not

(some) highly restrictive licenses and therefore the value of socialcohesion for A’s inter-project network for highly restrictive licensesis 0.5. If projects that use multiple licenses (such as B) are droppedfrom the sample, then project A’s inter-project network would con-sist of only project C and would erroneously be assigned a socialcohesion value for highly restrictive licenses of 0.

consider the social distance between a focal projectand other projects that started after it. As we explainbelow, values in this project-specific nearness matrixare used as a measure of social cohesion in the con-struction of a project-specific social influence variable.

5.4.2. Role Equivalence. In the OSS context,projects are role equivalent to the extent they employthe same technology platform in their development.OSS projects are assigned to foundries based on theircommon usage of a technology platform. Thus, weconstruct four versions of the variable, role equivalence,each indicating the fraction of projects in the focalproject’s foundry that are governed by the particularlicense type corresponding to the particular depen-dent variable. Consequently, the construction of thisvariable does not make use of the foundry-specificaffiliation networks in contrast to the social cohesionmeasure.

5.4.3. Susceptibility. We operationalize suscepti-bility to social influence with several measures. Tocapture the relative prior success of the licensor ofthe focal project (to test Hypothesis 4), we use theaverage size of his or her other projects at the incep-tion of the focal project at SF. The size of a projectis the number of developers involved in the project.An OSS project relies on voluntary contributions fromdevelopers to survive. A project that is able to attracta larger number of developers is more likely to sur-vive and achieve a stable release (Comino et al. 2007,Lerner and Tirole 2005b), which indicates a project’stechnical success (Grewal et al. 2006). Many studieshave used this measure to capture an OSS project’ssuccess (Comino et al. 2007, Lerner and Tirole 2005b,Singh et al. 2011a, Stewart et al. 2006). To test therobustness of our results involving this variable, weuse the project’s percentile rank on SourceForge asan alternative measure of project success (Lerner andTirole 2005b, Singh et al. 2011b). This measure cap-tures the ranking of the project based on its activityand popularity among developers and users and hasbeen used in prior research to capture OSS projectperformance (Lerner and Tirole 2005b). FollowingGreve (1998), we computed alternative measures ofthe social aspiration level for an OSS project as theaverage size and average rank of all projects in thesame foundry at the time of the focal project’s licensechoice. We then computed the relative success of anOSS project by subtracting the social aspiration levelfor the project from the project’s performance usingaverage size (RELSIZE) and project rank (RELRANK),respectively, at the time of the focal project’s licensechoice. These two variables are used to test H4.To capture the depth of experience of the licensor

of the focal project (Hypothesis 5), we use his or hertenure (TENURE) in the OSS environment, measured

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


as the number of months the licensor had been activeon SF at the time of license choice for the focal project,and experience (EXP), measured as the number ofprojects in which he or she had been involved by thestart of the focal project. To measure the diversity of alicensor’s experience with OSS licenses (DIVERSEXP)(Hypothesis 5), we use (Simpson 1949) index of diver-sity, also known as Blau’s index of heterogeneity andthe Hirschman-Herfindahl index. Experiential diver-sity was computed as a reverse-scaled Simpson index:1 É P

p2j , where pj is the proportion of licensor i’stotal months of experience across all projects spenton projects using license type j . We reverse codeSimpson’s measure so that higher values indicategreater experiential diversity.Our theory predicts the susceptibility variables will

reduce the influence of social cohesion and role equiv-alence on a focal project’s license choice. Thus, to esti-mate the effects of these variables on the susceptibilityof the focal project to social influence, we multiplythese measures with the two social proximity mea-sures. We include the susceptibility variables alone toensure proper model specification.

5.4.4. Infectiousness. We measure the infectious-ness of a prior adopter by its success (Hypothesis 3),where success is measured as the number of devel-opers (ASIZE) involved in the project at the starttime of the focal project. In the OSS community, thesuccess of a project reflects its social status in thecommunity (Stewart 2005). As an alternative mea-sure of the success of a prior adopter’s project weuse the project’s percentile rank on SourceForge. Weassess the robustness of our results to the use of thisalternative measure. Because infectiousness is a prop-erty of information generated by prior adopters, itsinfluence depends on whether or not it can reacha potential adopter and the distance it must travelto do so (Strang and Tuma 1993). Because the influ-ence of infectiousness depends on the social proxim-ity between a prior and potential adopter, we interactthe success measures with the two social proximitymeasures.8

8 Because there are many existing projects in a new project’sfoundry at the point in time it chooses a license, observations ofASIZE for existing projects need to be aggregated to the foundrylevel (so that each observation of a new project’s license choicehas a corresponding observation of ASIZE). In interacting eachexisting project’s ASIZE with its social proximity to a new (focal)project, we aggregate across all existing projects (that have adoptedlicense type j) by computing a network weight matrix that scalesthe ASIZE of each existing project (see §5.6.1). This approach doesnot control for the main effect of existing projects’ ASIZE. The con-struction of such a main effect variable would require similar aggre-gation across projects to the foundry level. Extant theory does notsuggest a particular approach to aggregation. In unreported results,we used the mean. We did this for both measures of infectiousness.

5.5. Control Variables

5.5.1. Structural Equivalence. To ensure theeffects of social cohesion and role equivalence are notconfounded by unobserved structural equivalence,we control for structural equivalence. Two actorsare structurally equivalent to the extent they haveties to the exact same other actors (Burt 1976).Projects in the same foundry are role equivalent,but not necessarily structurally equivalent. Structuralequivalence is measured by computing the Euclideandistance between projects i and j based on thedissimilarities in the relations of projects i and j toall other projects, k, in the same foundry. We follow(Burt 1987) and transform the Euclidean distances torepresent similarities in the relations of projects i andj by subtracting them from the maximum value ofEuclidian distance involving i. We compute structuralequivalence only for projects that are indirectly con-nected by at least one other project. Including projectsthat share a direct tie would make it impossible toseparate the effect of direct communication from thesocial comparison effect associated with structuralequivalence (Leenders 2002).

5.5.2. Focal Project Propensity. To minimize alter-native explanations and isolate the marginal effects ofthe explanatory variables, we include an extensive setof licensor- and project-specific controls.Licensor characteristics. Because of behavioral inertia,

a licensor who has experience with a particular licensetype from working on other projects may choose asimilar license type for their focal project. To controlfor this effect, we construct four versions of the vari-able EXPL, each indicating the fraction of projects towhich the focal licensor has contributed that are gov-erned by one of the four particular license types usedas our dependent variables.Project characteristics. Following Lerner and Tirole

(2005b), we construct an extensive list of softwareproject characteristics. Each project is characterizedby the following dimensions: intended audience (e.g.,end users, system administrators); topic (e.g., games,Internet); operating system (e.g., POSIX, Windows);user interface (e.g., GUI, text based); and natural lan-guage (e.g., English, Chinese). In all, we control for46 characteristics of a focal OSS project. For a list ofthese controls, see Tables A6–A9 in the online supple-ment. We control for every variable used by Lernerand Tirole (2005b) except project development stage.Because the license choice is made upon project regis-tration at SourceForge, all of our sample projects havea development stage of 0.

The inclusion of the two infectiousness variables as main effects inall estimated models do not change the results reported in Tables 3and 4.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


5.6. ModelThe basic framework for the license choice modeldeveloped here captures both economic and socialinfluence factors that affect a licensor’s choice of aparticular license type. Let yi be an indicator variablethat equals “1” if project i has license type L and “0”otherwise. The model is specified as

p4yi = 15= F 4ÇX+ÉZ+ íD5 and

p4yi = 15= e4ÇX+ÉZ+íD5

1+ e4ÇX+ÉZ+íD50

Here X represents the social influence variables, Zrepresents project characteristic controls, D representslicensor characteristic controls, and F is the logisticfunction. The variables in Z correspond to the eco-nomic incentives argument put forth by Lerner andTirole (2005b). Although the construction of Z and Dis straightforward, we explain the construction of Xin the next subsection.

5.6.1. Social Influence Variables. Social influencevariables account for three factors: social proximityof j to i, infectiousness of j , and the susceptibilityof i (Strang and Tuma 1993). We follow Burt (1987)and assume that concrete social proximity is subjec-tively perceived by a focal licensor. Thus, projectsat greater social distance may be perceived to haveless social influence than their actual social distancewould imply. In other words, social influence is per-ceived to exponentially decay with increasing socialdistance (Burt 1987). Given a concrete measure ofsocial proximity between a focal project, i, and theprojects in its social network, j , the focal project licen-sor’s subjective perception of that proximity can berepresented by the power function (proximity i to j)v(Burt 1987). The relative weight (or intensity) of socialinfluence perceived by a licensor from another projectis then computed as follows. Let P be the set ofall projects in the focal project’s foundry that startedbefore the focal project i, and Yp = 6y11y21 0 0 0 1yp] bea vector of indicator variables representing licensechoice of all projects in the set P . Whereas socialproximity captures the social frame of reference bywhich a licensor evaluates a type of license, the licensetype that socially proximal projects have previouslyadopted captures the normatively sanctioned arti-fact they are influencing the licensor to adopt. Inessence, in computing the social influence of a partic-ular license type on a particular licensor, the portionof projects in the licensor’s foundry that previouslyadopted the license type are weighted by their subjec-tive social proximity to the licensor. Finally, previousadopters of particular license types are also weightedin their social influence on a licensor by their infec-tiousness. Let wij capture the weight or intensity bywhich j influences i, relative to all other projects

in the network, moderated by the infectiousness ofproject j . These values, commonly referred to as net-work weights (Burt 1987, Leenders 2002), are calcu-lated as follows:

wij =4infectiousness of j⇥yj⇥social proximity of i to j5çP

k4infectiousness of k⇥social proximity of i to k5ç1

wi =X

j

wij1

where wi varies between zero and one. Higher valuesof wi correspond to higher social influence. The expo-nent v measures the extent to which focal project i isconservative in relying on others. Values of v muchlarger than one imply that a focal project’s evaluationof a license is affected by its most proximate projects,whereas fractional values indicate that it is affectedalmost by anyone to which it is connected. Hence, vdetermines the social frame of reference for the focalproject. This construction of wij allows us to accountfor multiple competing innovations. Social cohesionand role equivalence are two different measures ofsocial proximity that enter wi. As we explain belowin the results section, we identify and use the valueof v that maximizes the estimated likelihood of ourmodel. The susceptibility of a licensor to social influ-ence is estimated by multiplying wj with the licen-sor characteristics of project i expected to influencesusceptibility:

Xi =⇥wi1RELSIZEi ⇥wi1TENUREi ⇥wi1EXPi ⇥wi1

DIVERSEXPi ⇥wi

⇤0

5.6.2. Accounting for Unobserved Heterogeneity.Although we include an extensive number of con-trols, unobserved heterogeneity may still be a sourceof endogeneity and bias our parameter estimates. Forexample, licensors may have unobserved differencesin their intrinsic preferences for particular licensetypes or in their susceptibility to social influence.Licensors also may have unobserved differences intheir social relationships that influence their licensechoice, such as when some licensors have worked ontheir new projects with developers outside and/orprior to their registration on SF. We allow for unob-served heterogeneity in license choice behavior byallowing coefficients for social influence and licen-sor characteristics to vary randomly across licensors.Licensor random coefficients are introduced hierarchi-cally. The specified model is modified as

p4yi5= F 4ÇiXi +ÉZi + íiDi51 and

Çiíi =MVN4–1ËÇí51

where – is a vector that corresponds to the meanof

⇥Çiíi

⇤, and ËÇí is the corresponding covariance

matrix.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


5.6.3. Accounting for the Reflection Problem.Another source of endogeneity bias in estimating thesocial influence effects is the “reflection problem”(Manski 1993). Reflection poses a problem for causalidentification of social influence because (a) actorstypically form social ties with another based onhomophily—i.e., when they share similar attributes,beliefs, tastes, or interests (McPherson et al. 2001);(b) the variables that cause homophilous tie formationare typically unobserved by analysts investigating thecausal effect of one actor socially influencing another;and (c) the actors’ similar, yet unobserved attributes,beliefs, tastes, or interests are correlated with (orcause) their similarity of behavior. Consequently, theobserved effect of social influence via social ties maybe a spurious result of an omitted variables bias. Thereflection problem in our context would mean thelicensor of a new project chooses the same licensetype chosen by socially proximate projects in his orher social network not because of the treatment effectof social influence but because he or she is simi-lar in some unobserved ways (e.g., demographic orbehavioral) to the developers on those projects andit is this unobserved similarity that also causes thesesimilar actors to self-select into forming the socialrelationship connecting the projects. We address thissource of endogeneity in two ways. First, by construc-tion, only relationships and projects that exist priorto the adoption of a license by the focal project caninfluence the focal administrator’s license decision.This eliminates the potential for a simultaneity biasbecause the choice of license by a focal project tempo-rally follows the formation of relationships with otherprojects that previously adopted their own licenses.Second, we use an approach developed by (Bramoulleet al. 2009) to identify the true effect of social influ-ence by accounting for the reflection problem. Theapproach involves introducing unobservable effectscommon to all projects that belong to the same net-work component. The unobservable effects are treatedas component-specific fixed effects and eliminated byusing appropriate differencing.

5.6.4. Left-Censoring Bias. The possibility of hav-ing social ties with prior adopters of a license dependson the pool of adopters. Thus, the projects that startedat SourceForge close to its inception are likely to havefewer ties compared to projects that started later. Thiscreates a potential left-censoring bias. To minimizesuch a bias we used all projects to construct our socialnetworks, but we used only those projects that wereregistered from January 2002 to December 2003 to testour hypotheses.

5.6.5. Sample Selection Bias. Of the full sampleof 21,220 projects registered at SF during our study,15,913 projects were social isolates. The social influ-ence variables are undefined for such projects. Not

accounting for these projects in the estimation maylead to sample selection bias (Heckman 1976). Accord-ingly, we estimated a first-stage selection model forall 21,220 projects using instruments and then enteredthe inverse Mills ratio computed from this model inthe license choice models as a control for unobservedselection into the sample of socially connected newprojects. The first stage estimated the selection hazardas a function of all nonsocial influence variables alongwith two instruments: technical expertise of the licen-sor and network density of the focal project’s foundry.To measure technical expertise, we used the licensor’sself-reported level of expertise reported to SF. The cat-egory indicating the lowest level of expertise is “wantto learn,” which we coded as “1” if the developerselected that category and “0” otherwise. The want tolearn developers are more likely to be new to OSS andless likely to be part of the existing social networkof projects. Network density is the extent to which allthe projects within a foundry are connected with eachother. Higher density implies a higher level of inter-connection. We expect the likelihood a new projectis started by a developer from within the networkincreases with its density.

5.6.6. Estimation Procedure. The estimation pro-cedure consists of two steps. First, we estimateda selection model using probit regression. Second,we estimated four different license choice mod-els in which the inverse Mills ratio obtained fromstep one was included as a covariate. Each licensechoice model is estimated through a standard Markovchain Monte Carlo (MCMC) hierarchical Bayes logitestimation procedure, using a Gibbs sampler andthe Metropolis-Hastings algorithm coded in Matlab(Rossi et al. 2005). To reduce the autocorrelationbetween draws of the Metropolis-Hastings algorithmand to improve the mixing of the MCMC, we usedan adaptive Metropolis adjusted Langevin algorithm(Atchade 2006). In the hierarchical Bayes procedure,the first 100,000 observations were used as burn-in,and the last 25,000 were used to calculate the con-ditional posterior distributions. To assess the conver-gence of the MCMC, we compared the within- tobetween-variance for each parameter estimated acrossmultiple chains (Gelman and Rubin 1992).

6. ResultsDescriptive statistics for the key variables are reportedin Table 1. We log transformed TENURE and EXPbecause they were heavily skewed. We standardizedthe component variables (RELSIZE, TENURE, EXP,DIVERSEXP, social cohesion, and role equivalence). Theestimated model includes these standardized vari-ables and their interactions. This step reduced corre-lations to acceptable levels: variance inflation factors(VIFs) for all variables were below two.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Table 1 Descriptive Statistics (N = 51307 Observations)

Variables Mean Std. dev. Min Max

1 All licenses highly restrictive 4ALLHR 5 0055 005 0 12 Some licenses highly restrictive 4SHR 5 006 0049 0 13 All licenses restrictive 4ALLR 5 0073 0044 0 14 Some licenses restrictive 4SR 5 0076 0043 0 15 Social cohesion ALLHR 0054 0023 0 16 Social cohesion SHR 0064 0023 0 17 Social cohesion ALLR 0054 002 0 18 Social cohesion SR 0045 0025 0 19 Role equivalence ALLHR 0058 001 0 0076

10 Role equivalence SHR 0074 001 004 009611 Role equivalence ALLR 0077 0006 005 008912 Role equivalence SR 0087 0005 005 009813 Licensor’s experience 2094 1069 2 15

on OSS projects 4EXP 514 Average relative size of licensor’s 0013 1014 É3014 2094

prior projects 4RELSIZE 515 Licensor’s tenure in OSS 6019 6040 4 46

environment 4TENURE 516 Diversity of licensor’s experience 0065 0024 0 1

with OSS licenses 4DIVERSEXP 517 Structural equivalence ALLHR 0018 0026 0 118 Structural equivalence SHR 002 0028 0 119 Structural equivalence ALLR 0051 0048 0 120 Structural equivalence SR 0052 0048 0 1

The selection model results are not reported to con-serve space. Both technical experience and networkdensity significantly predicted the likelihood a newproject has ties to other projects of length greater thanone. Moreover, these variables were not correlatedwith any of the license choice variables.Before we explain the license choice results, we

need to explain the procedure that we followed forestimating the parameter ç, which is used to constructthe social influence variables. Following Burt (1987),we treat this parameter as a constant when calculatingthe value of social proximity. To identify the optimalvalue of ç, we compared likelihoods for different val-ues of ç (i.e., 0.1, 0.25, 0.5, 0.75, 1, 1.5, 2, 2.5, and 5)and then chose the value that provided the greatestlikelihood. Setting ç = 1 for both social cohesion andstructural equivalence provided the best likelihood.This indicates the licensor gives more weight to themost proximate prior adopters. We present results forsocial proximity measures calculated by using ç = 1.The results of the license choice model are provided

in Table 2. Coefficient posterior means and variancesare reported. Model 1 assesses the likelihood that alllicenses chosen by a focal project are highly restric-tive. Model 2 tests the probability that some of thelicenses are highly restrictive. Models 3 and 4 test theprobabilities for all and some licenses being restric-tive, respectively. To save space, software characteris-tics and time period effects, although estimated, arenot reported. Standardized coefficients are reportedfor all the models for comparison purposes.

Hypotheses 1 and 2 predicted a focal project’s socialproximity, in terms of social cohesion and role equiv-alence, to other OSS projects would increase the like-lihood it adopted the same license type as these otherprojects. Consistent with these hypotheses, the coef-ficients for social cohesion and role equivalence arepositive and significant in all models. Social cohesioneffect sizes are substantially larger than those for roleequivalence, which suggests that licensors are morestrongly influenced by direct communication thansymbolic communication in their inter-project socialnetworks. In Hypothesis 4, we predicted licensors thathad previously worked on relatively successful OSSprojects would be less susceptible to social influencein their license choice. Hypothesis 5 predicted licen-sors with greater depth and diversity of experienceon OSS projects would be less susceptible to socialinfluence. Consistent with these two hypotheses, theinteractions of the social proximity measures (socialcohesion and role equivalence) and susceptibility mea-sures are, with one exception, negative and signifi-cant across all models. The estimated coefficients forsocial cohesion, role equivalence, RELSIZE, TENURE,EXP, and DIVERSEXP are simple effects rather thanmain effects because the interaction terms are signifi-cant (Jaccard and Turrisi 2003). To assess the net effectof each of these variables, the main and the interac-tion effects must be combined. Using the results frommodel 1, the effect for social cohesion is

p = 610394É 00192 Social cohesion ⇤EXPÉ 00160 Social cohesion ⇤RELSIZEÉ 00291 Social cohesion ⇤TENURE

É 00077 Social cohesion ⇤DIVERSEXP70

The coefficient estimate of 1.394 for social cohe-sion in model 1 is conditional on RELSIZE, TENURE,EXP, and DIVERSEXP taking on the value of zero(thus removing the effects of the interactions involv-ing these variables), while the effect of social cohe-sion when RELSIZE, TENURE, EXP and DIVERSEXPare not zero will depend on their values and the val-ues of their coefficients. For example, when RELSIZE,EXP and DIVERSEXP are at their means, the effectof a one-unit change in social cohesion for one stan-dard deviation below and above the mean TENUREis 1.203 and 1.585, respectively.Hypothesis 3 predicted that prior adopters of OSS

licenses differ in their infectiousness based on theirsuccess and that the influence of socially proxi-mal projects on a focal project’s license choice willincrease with the success of these projects. To testthis prediction, we computed social influence weights(wi5 in two ways: (1) nonweighted social proximity(i.e., set infectiousness to one for all prior adopters)

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Table 2 Hierarchical Bayes Parameter Estimates for Logit License Choice Model

Model 1 Model 2 Model 3 Model 4

All Somelicenses licenses All Somehighly highly licenses licenses

restrictive? restrictive? restrictive? restrictive?

Coefficient Coefficient Coefficient CoefficientHypoth Variables posterior mean posterior mean posterior mean posterior mean

H1 Social cohesion 10394⇤⇤⇤ 10739⇤⇤⇤ 20144⇤⇤⇤ 10627⇤⇤⇤6003177 6003477 6004217 6003227

H2 Role equivalence 00966⇤⇤⇤ 10122⇤⇤⇤ 10446⇤⇤⇤ 10163⇤⇤⇤6003427 6002077 6001517 6002297

H4 Social cohesion⇥RELSIZE É00160⇤⇤⇤ É00134⇤ É00138⇤⇤⇤ É00097⇤6001057 6000457 6000967 6001017

H5 Social cohesion⇥ TENURE É00291⇤ É00216⇤ É00383⇤⇤ É00192⇤6000357 6000317 6000537 6000927

H5 Social cohesion⇥ EXP É00192⇤⇤ É00117⇤⇤⇤ É00103⇤⇤⇤ É00105⇤⇤⇤6002227 6002097 6003247 6002197

H5 Social cohesion⇥DIVERSEXP É00077⇤ É00065⇤ É00123⇤⇤ É00111⇤⇤6000327 6000277 6000967 6000557

H4 Role equivalence⇥RELSIZE É00145⇤⇤ É00097⇤⇤ É00101⇤⇤ É00077⇤6000267 6000057 6000177 6000267

H5 Role equivalence⇥ TENURE É00044⇤ É00079⇤⇤ É00117⇤⇤⇤ É00094⇤⇤6000137 6000127 6000237 6000037

H5 Role equivalence⇥ EXP É00034⇤ É00038⇤ É00026 É00029⇤6000147 6000197 6000197 6000117

H5 Role equivalence⇥DIVERSEXP É00119⇤⇤⇤ É00104⇤⇤⇤ É00181⇤⇤⇤ É00204⇤⇤⇤6000717 6000527 6000637 6001237

Licensor’s experience on OSS projects 4EXP5 É00009⇤ É00014⇤ É00055⇤⇤ É00024⇤⇤6000077 6000247 6000097 6000117

Relative size of licensor’s prior projects 4RELSIZE5 É00016⇤⇤ É00039⇤⇤⇤ É00027⇤⇤⇤ É00051⇤⇤⇤6000057 6000157 6000097 6000197

Licensor’s tenure at SourceForge 4TENURE5 É00512⇤⇤⇤ É00492⇤⇤⇤ É00447⇤⇤⇤ 00455⇤⇤⇤6003517 6002267 6001357 6000557

Diversity of licensor’s experience at SourceForge 4DIVERSEXP5 É00012 É00074 É00069 É000236001177 6001627 6001517 6001597

Structural equivalence 00177⇤⇤ 00284⇤⇤ 00573⇤⇤ 00443⇤⇤6002017 6002927 6002777 6001937

Number of projects 5,307 5,307 5,307 5,307

Notes. Posterior licensor-specific variances are given in square brackets below coefficient means; ⇤⇤⇤implies the 99% confidence interval does not include zero;⇤⇤implies the 95% confidence interval does not include zero; ⇤implies the 90% confidence interval does not include zero. All models include software charac-teristics and time dummy control variables. Social cohesion and role equivalence measures account for infectiousness based on size (ASIZE).

and (2) ASIZE-weighted social proximity (multipliedproximity by ASIZE of prior adopters). These arenonnested models. Following (Newton and Raftery1994), we compare these models on log marginal den-sity. The Bayesian information criterion (BIC), com-monly used for model selection in classical statisticalanalysis, asymptotically approximates the Bayesianposterior marginal density. Using this measure, thebest fitting model is the one that minimizes É2 logmarginal density. The models (each with a specificconstruction of wi5 were run separately and theirlog marginal densities were calculated. The É2 logmarginal density values are reported in Table 3. Forall four dependent variables, the models that account

Table 3 Choosing Competing Models 4É2 Log Marginal Density)

Role equivalence

Same Infectiousness DependentSocial cohesion infectiousness differ by size variable

Same infectiousness 121595006 121494073 ALLHR131691035 131407032 SHR141615029 141419005 ALLR141439050 141259094 SR

Infectiousness differ by size 121299093 121203024 ALLHR131187039 131095047 SHR141038043 131903097 ALLR131964020 131869003 SR

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


for infectiousness of prior adopters based on theirASIZE have the lowest É2 log marginal density. Thisindicates that more successful prior adopters are moreinfectious. This provides support for Hypothesis 3.The inclusion of foundry-level average ASIZE as amain effect does not change the reported results (seefootnote 10).Finally, by comparing the standardized coefficients

in Table 3, the major determinants of license choicecan be identified. The three most important character-istics that determine the type of license used for gov-erning an OSS project are the type of licenses used bythe closest projects in the social network of the licen-sor, the dominant license type in the project’s foundry,and the licensor’s tenure on OSS projects.We compared the predictive power of our full

model to that of the baseline controls-only model,which corresponds to Lerner and Tirole’s (2005b) eco-nomic model, to assess the incremental improvementin fit provided by our full model (see Table 4). Weused projects registered from Jan. 2002 to Dec. 2003as our estimation sample and projects registered fromJan. to Jun. 2004 as a prediction sample. We used theestimated coefficients from the two models to predictthe actual choice of license in the prediction sample.Because the adoption (or not) of a particular licensetype is binary, we assumed a particular license typewas correctly predicted when its predicted probabilityexceeded 50%. For both the baseline and full model,we calculated the overall hit rate, focal license hit rate,and nonfocal license hit rate.9 Such hit rates are stan-dard metrics of model fit used in prior research (e.g.,Netzer et al. 2008). We also computed the root meansquare prediction error (RMSPE) between the twomodels. The values for these metrics are presentedin Table 4. For all metrics our full model substan-tively outperforms the economic incentives model.Our results are robust to a number of robustnesschecks, which we report in the online supplement tothis paper.

9 The overall hit rate is defined as the percentage of projectsfor which the predicted license matches the actual outcome (e.g.,ALLHR and “not ALLHR”). For example, assuming there were 100projects in the prediction sample, 60 of which chose ALLHR and40 chose not ALLHR, and our model accurately predicted 70 ofthese licenses, then the overall hit rate is 70%. The focal license hitrate is calculated as the percentage of those projects’ licenses thatwere correctly predicted in the overall hit rate that were correctlypredicted as being a particular type of license (e.g., ALLHR). Con-tinuing the example, if out of the 70 projects accurately predictedthe model predicts 50 ALLHR adopter projects correctly, then thefocal license hit rate for ALLHR is 83.33% (50/60). The nonfocallicense hit rate is the percentage of those projects’ licenses thatwere correctly predicted in the overall hit rate that were correctlypredicted as being not a particular type of license (not ALLHR).In the example, of the 40 not ALLHR projects in the sample, themodel accurately predicts 20 (70 É 50), which implies a nonfocallicense hit rate of 50% (20/40).

Table 4 Model Fitness and Predictive Power Analysis

Social influence+Economic Economic

incentives model incentives modelMeasures 4%5 4%5

Overall license hit rate (ALLHR) 42039 71072Focal license hit rate (ALLHR) 45021 75059Nonfocal license hit rate (ALLHR) 38094 66099RMPSE (ALLHR) 7042 3093Overall license hit rate (SHR) 47084 73061Focal license hit rate (SHR) 49005 78024Nonfocal license hit rate (SHR) 46002 66067RMSPE (SHR) 6091 3066Overall license hit rate (ALLR) 63095 82037Focal license hit rate (ALLR) 74044 84029Nonfocal license hit rate (ALLR) 35059 77018RMSPE (ALLR) 6058 3016Overall license hit rate (SR) 65052 85019Focal license hit rate (SR) 76017 87092Nonfocal license hit rate (SR) 31079 76054RMSPE (SR) 5096 2091

7. DiscussionThis study was motivated by gaps in the literature oninnovation adoption and diffusion, particularly thatwhich employs the heterogeneous diffusion frame-work, and the OSS licensing literature. First, existinginnovation adoption and diffusion research provideslittle insight into how, when, or why social influenceaffects the adoption and diffusion of competing arti-facts (Strang and Soule 1998). Second, this researchdoes not consider when or how the experiences oforganizational members who have worked with par-ticular innovations in their previous employers affecttheir current organizations’ adoption of such innova-tions. Finally, research provides an incomplete under-standing of OSS license choice because it ignores thepotential social influence of prior adopters of partic-ular licenses on a project manager’s license choiceand how this influence may vary over time and by aproject manager’s social proximity to established OSSprojects, characteristics of these projects, and the man-ager’s previous OSS experience.In addressing these limitations, we examined the

conditions under which prior adopters of competingOSS licenses socially influence how a new OSS projectchooses among such licenses and how the experiencesof the project manager of a new OSS project withparticular competing licenses affects its susceptibil-ity to this social influence. In investigating this ques-tion, we adapted the heterogeneous diffusion model(Strang and Tuma 1993) to accommodate multiple,competing innovations and extended it by identifyinga novel source of influence on a potential organiza-tional adopter’s susceptibility to the social influenceof prior adopters. This allowed us to examine howthe decision to adopt a particular type of OSS license

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


is shaped not only by economic incentives but also bythe license choices made by other projects to whicha licensor is socially proximate. We tested our pre-dictions in a sample of 5,307 OSS projects hosted atSourceForge.After controlling for a large variety of factors shown

to affect OSS license choice (Lerner and Tirole 2005b),our findings reveal the most important factor deter-mining a new project’s license choice is the type oflicense chosen by existing projects that are closer toit in its inter-project social network. Moreover, thelikelihood that a new OSS project adopts a particu-lar license increases when more role equivalent OSSprojects have previously adopted such a license andwhen these projects are large and successful. We alsofound that project managers with greater depth anddiversity of experience in the OSS community andwho have previously been members of successful OSSprojects are less susceptible to social influence. Finally,the results indicate that social influence dominateseconomic incentives in the choice of license type.These results, which are robust to the use of a largenumber of controls, substantial efforts to control forvarious sources of endogeneity, alternative measuresof the dependent variable and alternative estimationtechniques, have important implications for researchand practice.

7.1. Implications for ResearchThis study has substantive implications for researchon: innovation adoption and diffusion informed bythe heterogeneous diffusion framework, open sourcesoftware licensing, and the governance of economicexchange. First, the results contribute to innovationadoption and diffusion research by showing howsocial influence from prior adopters affects a poten-tial adopter’s choice among competing innovations.Although nearly all innovation adoption and diffu-sion research restricts its focus to a single innovation,potential adopters are frequently confronted withchoosing among multiple, competing innovations. Weshow that the heterogeneous diffusion framework canbe readily extended to the context of competing inno-vations by focusing on how the prior adoption ofone particular innovation affects a potential adopter’schoice among all competing innovations, and howthis relationship is influenced by characteristics of theprior adopters and potential adopters and the socialproximity between them. The results generated fromthis approach suggest early adoption choices of oneparticular innovation can stimulate subsequent adop-tions of the same innovation, potentially resulting in abandwagon effect in which one (or a few) innovationsdiffuse widely and come to dominate while others arelargely locked out (Abrahamson and Rosenkopf 1997).Our results suggest OSS licenses are subject to such

a positive feedback effect in their adoption and diffu-sion and such effects are more likely to occur when(a) early adopters of a particular innovation are moreinfectious (e.g., when they are viewed as successfulor prestigious), (b) prior and potential adopters aremore socially proximate, and (c) potential adoptersare more susceptible to social influence (e.g., whenthey lack direct experience with any of the competinginnovations). In contrast, adoption and diffusion in apopulation of actors is less likely to tip in favor of one(or a few) of many competing innovations when thesefactors are absent. The results of this study providenovel insight into understanding when a populationof actors will exhibit more or less heterogeneity inthe adoption and diffusion of competing innovationsbased on social influence.Our results also contribute to innovation adoption

and diffusion research by showing how an organi-zation’s susceptibility to the social influence of prioradopters is shaped by the experiences of organiza-tional members who have worked with particularinnovations in other organizations. Although existingresearch assumes potential organizational adopters donot have direct experience with an innovation beforeadopting it, interorganizational employee mobilitymakes it possible for employees to gain direct expe-rience with an innovation in one organization andtransfer this experience when they move to a neworganization. In the OSS context, we showed thatnew projects founded by managers with little experi-ence with other OSS projects or who had contributedto unsuccessful OSS projects were most influencedby prior adopters in their inter-project social net-works. With more, and more diverse, experience andgreater success on prior projects licensors becameless susceptible to social influence. These results sug-gest the experiences of a key organizational decisionmaker who has worked with particular innovationsin other organizations can affect his or her currentorganization’s susceptibility to the social influence ofprior adopters’ choices. These results contribute toan understanding of the micro-foundations of orga-nizational innovation adoption because they showhow organizational adoption decisions are affectedby the experiences of individual organizational deci-sion makers. Little research has examined the micro-foundations of the organizational susceptibility to thesocial influence of prior adopters (Wejnert 2002).This study also has clear implications for research

on OSS license choice. The little research that hasexamined this topic has not considered the potentialfor established OSS projects to socially influence thelicense choice of new projects, which is surprisinggiven the substantial uncertainty and confusion licen-sor typically face in choosing an appropriate license.Our results suggest social influence from prior license

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


adopters strongly affects the scope of license cho-sen when licensors are uncertain about the properlicense choice. Given that the effect sizes of the socialinfluence variables were substantially greater than allother variables, our results suggest that prior researchprovides a substantively incomplete understanding ofOSS license choice. Our theory and results do notoffer an alternative explanation of OSS license choiceas we do show economic incentives matter, but we doprovide an important, and heretofore missing, com-plement to existing research by demonstrating theimportant role social influence plays. Both economicincentives and social influence from socially proxi-mate established projects affect OSS license choice.Our results also indicate that the effect of socialinfluence on license choice declines, allowing for astronger influence of economic incentives, as a licen-sor’s depth and diversity of experience on prior OSSprojects and the success of these projects increases.Beyond providing an improved understanding of

the OSS license choice, our results have implicationsfor understanding the origins and influence of thesocial institutions that govern economic exchange. Weaddress a fundamental question at the intersectionof economics and sociology of how societies arriveat particular social institutions that govern economicactivity (North 1990, Zukin and DiMaggio 1990). Ourresults indicate the development of social institutionsthat order economic activity, such as particular con-tractual governance mechanisms (North 1990), arethe result of a social construction process by andamong organizations in which social networks playa fundamental role. Under conditions of uncertaintyabout the appropriate contractual governance prac-tice to adopt, socially proximate and infectious sourceorganizations serve as influential social referents forpotential adopters, resulting in a process of selec-tive imitation that has a homogenizing effect on thetypes of contractual governance adopted. This insightconfirms previous research that suggests particularpractices diffuse and become widely shared and takenfor granted (i.e., institutionalized) in particular orga-nizational fields through a process of mimetic isomor-phism among socially proximate organizations (Davisand Greve 1997, DiMaggio and Powell 1983, Zucker1987). Our results also indicate there are limits to thisprocess of mimetic isomorphism in that adopters withmore depth and diversity of experience with particu-lar governance practices are less likely to conform tothe governance choices of socially proximate sourceorganizations. In elucidating the role of networksin the diffusion and institutionalization of particulargovernance practices, we provide a meso-level expla-nation by bridging macro-level explanations, such asthose that highlight the influence of the frequencyof prior adoption, with micro-level explanations that

focus on individual utility-maximizing behavior. Wewere able to bridge these perspectives theoreticallyand methodologically by adapting and extending theheterogeneous diffusion model.

7.2. Implications for PracticeOur results also have implications for practice. Uncer-tainty about OSS licenses seems to be driving manymanagers of new projects to seek guidance fromlicense choices made by existing projects. Indeed,prior research (Rosenberg 2000) and our observationsof queries posted by open source software develop-ers to Internet-based discussion forums reinforce theinsight that managers of new OSS projects face sub-stantial uncertainty when choosing a license. Researchthat examines the role of social influence in adop-tion decisions characterized by substantial uncer-tainty prescribes centralized, third-party systems togather and disseminate information to decision mak-ers (Bikhchandani et al. 1998). Accordingly, we recom-mend that centralized educational initiatives shouldbe established to translate the results of research onOSS license choice and disseminate it to would-beproject managers to help them choose an appropriatelicense for their projects. One useful form that suchinitiatives could take would be the development andimplementation of an interactive software tool thatwould allow potential licensors to specify character-istics of their projects and receive guidance on thebest license options. Organizations that might be wellsuited for this role include the Free Software Founda-tion, the Open Source Initiative and SourceForge.net.The results of this study also provide insight into

how to stimulate the adoption and diffusion of acompeting innovation. First, given the positive influ-ence of the number of prior adopters on a poten-tial adopter’s choice, innovation promotion strategiesshould focus on establishing a critical mass of earlyadopters. These efforts should be targeted at seedingthe innovation with highly prestigious or successfulactors as our results suggest their adoption decisionsare particularly influential in subsequent adoption.Because such actors also tend to be less susceptibleto social influence themselves, marketing strategiesto encourage adoption will need to be particularlypersuasive and attractive. Next, given the positiveinfluence of the social proximity of prior adopters onpotential adopters, promotional strategies should alsofocus on seeding the innovation with actors who havehigh closeness centrality in their respective networksand who have high role equivalence with many otheractors. Finally, because we show that actors who lackexperience with any of the competing innovations arethe most susceptible to social influence, promotionalstrategies should target such actors and emphasize inpromotional communications to them the number of

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


prior adopters and those that are particularly presti-gious and successful.

7.3. LimitationsThe limitations of this study suggest opportunities forfuture research. First, given the motivation for thisstudy, we focused on aspects of existing and newprojects and the network ties among them. Conse-quently, we did not consider the potential influenceof information sources beyond the population of OSSprojects. Innovation diffusion research suggests exter-nal actors such as the mass media, consultants, andprofessional communities influence innovation adop-tion and diffusion by shaping how potential adoptersin a population interpret innovations as appropriateand legitimate (Strang and Soule 1998, Wejnert 2002).The question of how OSS license choice is influencedby these external actors is thus an important questionfor future research. Second, although we argued thatinformation transmission, observation, and learningare the causal mechanisms underlying social influ-ence, like most network diffusion studies our datado not allow us to directly observe and distinguishamong these explanations. Future research shouldtherefore seek to isolate these mechanisms, possiblythrough longitudinal qualitative study, in order to bet-ter specify how and why social influence from prioradopters affects the choices made by new adopters.A final limitation concerns the generalizability of theresults as our results may be limited to the timeperiod analyzed. The time period we studied includesthe founding of SourceForge and thus an early era incentrally hosted OSS projects, which may have facedgreater uncertainty about license choice than projectsstarted much later. Thus, additional data from morerecent time periods may be needed to externally val-idate our results.

Electronic CompanionAn electronic companion to this paper is available aspart of the online version at http://dx.doi.org/10.1287/isre.1120.0449.

AcknowledgmentsThe authors would like to thank Henrich Greve and SonaliShah for comments on an earlier draft of this manuscript.Both authors contributed equally to this paper.

ReferencesAbrahamson E, Rosenkopf L (1997) Social network effects on the

extent of innovation diffusion: A computer simulation. Organ.Sci. 8(3):289–309.

Ansari SM, Fiss PC, Zajac EJ (2010) Made to fit: How practices varyas they diffuse. Acad. Management Rev. 35(1):67–92.

Argyris C, Schön DA (1974) Theory in Practice: Increasing ProfessionalEffectiveness, 1st ed. (Jossey-Bass Publishers, San Francisco).

Atchade Y (2006) An adaptive version for the Metropolis adjustedLangevin algorithm with a truncated drift. Methodology Com-put. Appl. Probab. 8(2):235–254.

Audia PG, Goncalo JA (2007) Past success and creativity over time:A study of inventors in the hard disk drive industry. Manage-ment Sci. 53(1):1–15.

Bandura A (1986) Social Foundations of Thought and Action (Prentice-Hall, Englewood Cliffs, NJ).

Bikhchandani S, Hirshleifer D, Welch I (1998) Learning from thebehavior of others: Conformity, fads, and informational cas-cades. J. Econom. Perspect. 12(3):151–170.

Bothner M (2003) Competition and influence: The diffusion of thesixth-generation processor in the global computer industry.Amer. J. Sociol. 108(6):1175–1210.

Bramoulle Y, Djebbari H, Fortin B (2009) Identification of peer effectsocial networks. J. Econometrics 150(1):41–55.

Burt R (1976) Positions in networks. Soc. Forces 55(1):93–122.Burt RS (1982) Toward a Structural Theory of Action (Academic Press,

New York).Burt RS (1987) Social contagion and innovation: Cohesion versus

structural equivalence. Amer. J. Sociol. 92(6):1287–1335.Cattani G, Ferriani S (2008) A core/periphery perspective on indi-

vidual creative performance: Social networks and cinematicachievements in the Hollywood film industry. Organ. Sci.19(6):824–844.

Coleman J, Katz E, Menzel H (1966) Medical Innovation (BobbsMerrill, New York).

Comino S, Manenti M, Parisi M (2007) From planning to mature:On the success of OSS projects. Res. Policy 36(10):1575–1586.

Cyert R, March J (1963) Behavioral Theory of the Firm (Blackwell,Oxford, UK).

Davis GF, Greve HR (1997) Corporate elite networks and gover-nance changes in the 1980s. Amer. J. Sociol. 103(1):1–37.

de Laat PB (2005) Copyright or copyleft?: An analysis of propertyregimes for software development. Res. Policy 34(10):1511–1532.

Demil B, Lecocq X (2006) Neither market nor hierarchy nor net-work: The emergence of bazaar governance. Organ. Stud.27(10):1447–1466.

DiMaggio P, Powell W (1983) The iron cage revisited: Institutionalisomorphism and collective rationality in organized fields.Amer. Sociol. Rev. 48(2):147–160.

Festinger L (1954) A theory of social comparison processes. HumanRelations 7(2):117–140.

Fichman RG (2004) Real options and IT platform adoption: Implica-tions for theory and practice. Inform. Systems Res. 15(2):132–154.

Flingstein N (1985) The spread of multidivisional form among largefirms, 1919–1979. Amer. Sociol. Rev. (50):377–391.

Gallaugher JM, Wang YM (2002) Understanding network effectsin software markets: Evidence from Web server pricing. MISQuart. 26(4):303–327.

Gelman A, Rubin D (1992) Inference from iterative simulation usingmultiple sequences. Statist. Sci. 7(4):457–472.

Greve H (1998) Performance, aspirations, and risky organizationalchange. Admin. Sci. Quart. 44(1):58–86.

Greve H (2005) Interorganizational learning and heterogeneoussocial structure. Organ. Stud. 26(7):1025–1049.

Grewal R, Lilien G, Mallapragada G (2006) Location, location, loca-tion: How network embeddedness affects project success inopen source systems. Management Sci. 52(7):1043–1056.

Guler I, Guillén MF, Macpherson JM (2002) Global competition,institutions, and the diffusion of organizational practices: Theinternational spread of ISO 9000 quality certificates. Admin. Sci.Quart. (2):207–232.

Hahn J, Moon J, Zhang C (2008) Emergence of new project teamsfrom open source software development networks: Impact ofprior collaboration ties. Inform. Systems Res. 19(3):369–391.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Hall BH (2004) Innovation and diffusion. Fagerberg J, Mowery DC,Nelson RR, eds. Oxford Handbook of Innovation (Oxford Univer-sity Press, Oxford, UK), 459–485.

Hardgrave B, Davis F, Riemenschneider C (2003) Investigatingdeterminants of software developers’ intentions to followmethodologies. J. MIS 20(1):123–151.

Haunschild P (1994) How much is that company worth?: Interor-ganizational relationships, uncertainty, and acquisition premi-ums. Admin. Sci. Quart. 30(3):391–411.

Heckman JJ (1976) Common structure of statistical-models of trun-cation, sample selection and limited dependent variables anda simple estimator for such models. Ann. Econom. Soc. Measure-ment 5(4):475–492.

Henrich J, Gil-White FJ (2001) The evolution of prestige: Freelyconferred deference as a mechanism for enhancing the ben-efits of cultural transmission. Evolution and Human Behav.22(3):165–196.

Homans G (1961) Social Behavior: Its Elementary Forms (Hardcourt,Brace and World, New York).

Jaccard J, Turrisi R (2003) Interaction Effects in Multiple Regression(Sage, Thousand Oaks, CA).

Janis IL (1972) Victims of Groupthink; A Psychological Study of Foreign-Policy Decisions and Fiascoes (Houghton, Boston).

Jeyaraj A, Rottman J, Lacity M (2006) A review of the predic-tors, linkages, and biases in IT innovation adoption research.J. Inform. Tech. 21(1):1–23.

Kahneman D, Slovic P, Tversky A (1982) Judgment Under Uncer-tainty: Heuristics and Biases (Cambridge University Press,Cambridge, New York).

Kraatz MS (1998) Learning by association? Interorganizational net-works and adaptation to environmental change. Acad. Manage-ment J. 41(6):621–643.

Krishnamurthy S (2002) Cave or community? An empirical exami-nation of 100 mature open source projects. First Monday 7(6),http://firstmonday.org/htbin/cgiwrap/bin/ojs/index.php/fm/rt/printerFriendly/1477/1392.

Laumann E, Marsden P, Prensky D (1983) The boundary specifi-cation problem in network analysis. Burt RS, Minor MJ, eds.Applied Network Analysis (Sage, Beverly Hills, CA), 18–34.

Leenders R (2002) Modeling social influence through networkautocorrelation: Constructing the weight matrix. Soc. Networks24(1):21–47.

Lerner J, Tirole J (2002) Some simple economics of open source.J. Indust. Econom. 50(2):197–234.

Lerner J, Tirole J (2005b) The scope of open source licensing. J. Law,Econom., Organ. 21(1):20–56.

Louis M (1980) Surprise and sense making: What newcomers expe-rience in entering unfamiliar organizational settings. Admin.Sci. Quart. 25(2):226–251.

Manski C (1993) Identification of endogenous social effects: Thereflection problem. Rev. Econom. Stud. 60(3):531–542.

March JG, Simon HA (1958) Organizations (Wiley, New York).Marsden P (2005) Recent developments in network measurement.

Carrington P, Scott J, Wasserman S, eds. Models and Meth-ods in Social Network Analysis (Cambridge University Press,New York), 8–30.

Marsden P, Friedkin N (1993) Network studies of social influence.Sociol. Methods Res. 22(1):(127–151).

McGowan D (2001) Legal implications of open source software.Univ. Illinois Law Rev. 10(3):241–304.

McPherson M, Smith-Lovin L, Cook J (2001) Birds of a feather:Homophily in social networks. Amer. Rev. Sociol. 27:415–444.

Mizruchi MS (1993) Cohesion, equivalence, and similarity of behav-ior: A theoretical and empirical assessment. Soc. Networks15(3):275–307.

Netzer O, Latin J, Srinivasan V (2008) A hidden Markov model ofcustomer relationship dynamics. Marketing Sci. 27(2):185–204.

Newton M, Raftery A (1994) Approximate Bayesian inferenceby the weighted likelihood bootstrap. J. Royal Statist. Soc. B56(1):3–48.

Niiranen O (2009) Licensing 101 for Open Source Projects: Choos-ing an Open Source License. Linux.com, http://www.linux.com/news/biz-os/legal/26690-licensing-101-for-open-source-projects-choosing-an-open-source-license.

North DC (1990) Institutions, Institutional Change, and Economic Per-formance (Cambridge University Press, Cambridge, New York).

Phelps C, Heidl R, Wadhwa A (2012) Knowledge, networks, andknowledge networks: A review and research agenda. J. Man-agement 38(4):1115–1166.

Podolny J (2001) Networks as the pipes and prisms of the market.Amer. J. Sociol. 107(1):33–60.

Rogers E (2003) Diffusion of Innovations (Free Press, New York).Rogers E, Kincaid D (1981) Communication Networks: Towards a New

Paradigm for Research (Free Press, New York).Rosenberg DK (2000) Open Source: The Unauthorized White Papers

(John Wiley & Sons, Hoboken, NJ).Rossi P, Allenby G, McCulloch R (2005) Bayesian Statistics and Mar-

keting (John Wiley & Sons, Hoboken, NJ).Schneider SL (1992) Framing and conflict—Aspiration level

contingency, the status-quo, and current theories of riskychoice. J. Experiment. Psych.-Learn. Memory and Cognition18(5):1040–1057.

Sen R, Subramaniam C, Nelson M (2009) Determinants of the choiceof open source software license. J. Management Inform. Systems25(1):207–240.

Shah S (2006) Motivation, governance, and the viability of hybridforms in open source software development. Management Sci.52(7):1000–1014.

Shannon C (1949) Communication in the presence of noise. Proc.IRE 37(1):10–21.

Shultz A (1964) Studies in Social Theory Brodersen A, ed.(The Hague, Martinus Nijhoff, Leiden, The Netherlands), 8–31.

Simcoe TS, Waguespack DM (2011) Status, quality, and attention:What’s in a (missing) name? Management Sci. 57(2):274–290.

Simpson EH (1949) Measurement of diversity. Nature 163:688.Singh PV (2010) The small world effect: The influence of macro

level properties of developer collaboration networks on opensource project success. ACM Trans. Software Engrg. Methodology20(2):1–27.

Singh PV, Tan Y (2010) Developer heterogeneity and formation ofcommunications networks in open source software projects.J. Management Inform. Systems 27(3):179–210.

Singh PV, Tan Y, Mookerjee V (2011a) Network effects: The influ-ence of structural social capital on open source software projectsuccess. Management Inform. Systems Quart. 35(4):813–829.

Singh PV, Tan Y, Youn N (2011b) A hidden Markov model ofdeveloper learning dynamics in open source software projects.Inform. Systems Res. 22(4):790–807.

Song J, Almeida P, Wu G (2003) Learning-by-hiring: When is mobil-ity more likely to facilitate interfirm knowledge transfer? Man-agement Sci. 49(4):351–365.

Stallman R, Lessing L, (2002) Free software, free society: Selected essaysof Richard M. Stallman (Free Software Foundation, Boston, MA).

Stewart D (2005) Social status in an open-source community. Amer.Sociol. Rev. 70:823–842.

Stewart K, Ammeter T, Likoebe M (2006) Impacts of license choiceand organizational sponsorship on user interest and develop-ment activity in open source software projects. Inform. SystemsRes. 17(2):126–144.

Strang D, Soule S (1998) Diffusion in organizations and socialmovements: From hybrid corn to poison pills. Annual Rev.Sociol. 24(5):265–290.

Strang D, Tuma N (1993) Spatial and temporal heterogeneity indiffusion. Amer. J. Sociol. 99(3):614–639.

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Susarla A, Oh J, Tan Y (2012) Social networks and the diffusion ofuser-generated content: Evidence from YouTube. Inform. Sys-tems Res. 23(1):23–41.

Uzzi B, Spiro J (2005) Collaboration and creativity: The small worldproblem. Amer. J. Sociol. 11(2):447–504.

Wasserman S, Faust K (1994) Social Network Analysis (CambridgeUniversity Press, London).

Wejnert B (2002) Integrating models of diffusion of innovations:A conceptual framework. Annual Rev. Sociol. 28:297–326.

Winship C, Mandel M (1983) Roles and positions: A critique andextension of the blockmodeling approach. Sociol. Methodology1984:314–344.

Zucker LG (1987) Institutional theories of organization. Annual Rev.Sociol. 13:443–464.

Zukin S, DiMaggio P (1990) Structures of Capital: The Social Organi-zation of the Economy (Cambridge University Press, Cambridge[England], New York).

Cop

yright:INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.

Mack Institute for Innovation Management - Networks, Social … · 2020-01-06 · Singh and Phelps:...

Documents

Transcript of Mack Institute for Innovation Management - Networks, Social … · 2020-01-06 · Singh and Phelps:...